Dataflow Monitoring

Dataflow Monitoring

Nicoletta Garelli

ALICE, ATLAS, CMS & LHCb joint workshop on DAQ@LHC

12-14 March 2013Château de Bossey

What Dataflow Monitoring Means

• Monitoring in real time the flow of data to ensure optimal data taking– from detector readout to permanent storage– trigger & DAQ quantities (counters, data rate, buffer

occupancy, etc.)• Avoid dead-time – and eventually allow to fix problems

• Each experiment uses its own jargon to indicate the same thing

12/03/2013 N. Garelli (SLAC) - Dataflow Monitoring 2

Requirements

• Access any relevant information in real time to follow data taking• Online aggregation & data correlation• Online problem detection: dead-time,

data losses, etc.• Archive: access historical data for

diagnostics, statistics, post-mortem• Use monitoring data to trigger alarms

and/or automatic actions to recover problems


Basics

Added later

Evolution• Users: shifters & experts• LHC operations …at the beginning– scattered information and rudimentary tools– shifters: intense monitoring activity – experts: high presence in control room + ringing on-call phone


• LHC operations …routine– coherent information and optimized tools– automate as much as possible to reduce shifter’s

tasks• see Luca’s talk of this morning

– move from custom GUI to ubiquitous web based tools• let’s do all with a smartphone

THE 4 ARCHITECTURES


Middleware

• Each experiment developed 4 different DAQ systems using different technologies• Variety reflected in dataflow monitoring middleware– LHCb & ALICE: Distributed Information Management (DIM)

• client/server paradigm, light weight

– ATLAS: Information Service (IS)• custom library on top of CORBA• client-server communication model where information is stored

in memory by so called IS servers

– CMS: Web Service• Cross-DAQ (XDAQ) framework


ALICE DAQ


• ~300 processes on ~300 machines• 100k dataflow information published every 5 s ~3 GB/h

NOTE:HLT monitoring & dedicated expert storage monitoringnot discussed here

ALICE Dataflow Monitoring Architecture

• “Logbook”– much more

than what you think

– PHP12/03/2013 N. Garelli (SLAC) - Dataflow Monitoring 8

DAQ processes

DIM / SMI

Status Display Backpressure Monitor Logbook

MySQL

• Based on DIM/SMI– SMI: framework for designing

and implementing distributed control systems developed by DELPHI

• MySQL:– store system

configuration – LDC&GDC write

run info – Archive.

• Logbook as visualization

• 2 monitoring applications

• Tcl/Tk

ALICE Visualization


ALICE & Android


ATLAS DAQ

• O(20k) processes on ~2k machines• 1M dataflow

information published every 5-10 s ~4 GB/h


Custom HW

Processing UnitProcessing UnitProcessing UnitProcessing UnitProcessing Node

Level 2

Processing UnitProcessing UnitProcessing UnitProcessing UnitProcessing Node

Event Filter

Other

FE FE FE

Calo/ Muon

ROD ROD ROD

Det

ecto

r Rea

dout

Processing UnitProcessing UnitReadOut System

Processing UnitProcessing UnitEvent Builder

Processing UnitData Logger

DC

BE

Level 1

~150

~100

~5

~8k

~8k

ATLAS Dataflow Monitoring Architecture


Information Service (IS)

~100 servers

Monitoring GUI

applicationsWeb Server

(Apache)

Web IS (Python CGI)

Web Browser

DAQ processes

Mirror Information Service

Monitoring GUI applications

CERN GPN

• WEB IS: IS gives information access on demand via HTTP protocol‐ python wrapper accepts

HTTP requests & sends back dynamically formed XML text (value of IS obj pointed by given URL)

• Mirror IS: real-time copy from IS (ATCN) to mirror counterpart in CERN GPN

• Archive: None.‐ information stored &

accessed for ~2 month in RDD files each ~30 s via network monitoring system

Web Server(Apache)

Web IS (Python CGI)

Web Browser

Shifter’s Tools in 2012

• DAQ Panel: tool portal for shifters• DFSummary– dynamically constructed web page

which computes & displays most important dataflow parameters (~200 variables)

– ~30 s update rate• Busy Panel: Qt application for

monitoring dead-time• Shifter Assistant– see Luca’s talk of this morning


CMS DAQ


• O(20k) processes on ~2k machines• O(600k) dataflow information published every 1-5 s ~8 GB/h

XDAQ Monitoring & Alarming Service• Fully scalable distributed

monitoring & alarming system• Service-oriented architecture

organized in 3-tier structured collection of communicating components:– Sensor: report monitoring data – Eventing: scalable publisher-

subscriber service orchestrated by a load balancer application (Broker)

– Collector: build relational tables– Live Access Service:

presentation of raw data (Web Service)


• Archive: automatic persistency of collected tables in ORACLE according to configuration– Subset of information stored: ~30 GB/y

Monitoring as a Service• XDAQ as a Service (XaaS): common infrastructure for both central DAQ &

sub-detectors


...

ecal

hcal

csc

xaasXDAQ as a service

cdaq

sentinel monitor application

central xaas

• interoperable services providing standard functionalities for use in XDAQ environment

• All processes organized into searchable groups known as zones

• zone defines scope of a distributed XDAQ application

• Each zone has its own monitoring data types (flashlists)

CMS Visualization• LabView DAQMon


LHCb DAQ

• O(40 k) processes on 2k machines• 4M dataflow information published

every 5s ~ 11.5 GB/h12/03/2013 N. Garelli (SLAC) - Dataflow Monitoring 18

SWITCH

HLT farm

Detector

TFC System

SWITCHSWITCH SWITCH SWITCH SWITCH SWITCH

READOUT NETWORK

L0 triggerLHC clock

MEP Request

Event building

Front-End

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

CPU

Readout Board

Expe

rimen

t Con

trol

Sys

tem

(EC

S)

VELO ST OT RICH ECal HCal MuonL0

Trigger

Event dataTiming and Fast Control SignalsControl and Monitoring data

SWITCH

MON farm

CPU

CPU

CPU

CPU

Readout Board

Readout Board

Readout Board

Readout Board

Readout Board

Readout Board

FEElectronics

FEElectronics

FEElectronics

FEElectronics

FEElectronics

FEElectronics

FEElectronics

LHCb Dataflow Monitoring Architecture


UI

Archive

AlarmUI

UIs

UIDetailedUIs

StatusCountersRatesAlarms

DetailedCounters

ControlSystem

UIReadoutUnits

UITiming &Trigger

CountersRates

CountersRates

Infrastr

Dataflow

DQHisto

HLT

Infrastr

Dataflow

DQHisto

Monitoring

Infrastr

Dataflow

DQHisto

Reconstr.

Histos

Histos

Histos

HLT SubFarms X 56

DIM

Infrastr

Dataflow

DQHisto

Infrastr

Dataflow

DQHisto

HLT Nodes X ~30DIM

DIM

HLT SF Ctrl

Infrastr

Dataflow

Storage

EventData

Status,Counters

Counters

Automatic Actions (aggregation)(aggregation)

Archive: • PVSS ORACLE

DB• O(200k) values

every 5s

LHCb Visualization

PVSS GUI


VT100 graphicsdetailed UI

CONCLUSIONS


Satisfied?“YES, it does the job”

“ … BUT …”– 4 different solutions for the same problem …– sharing experience and maybe even future common

solutions?

Luciano’s talk on Thursday


Dataflow Monitoring

Documents

data rate

flow of data

data losses

postmortemuse monitoring

beginningscattered information

relevant information

access historical data

real time